Skip to content

Commit b7a3ca2

Browse files
authored
Merge pull request #9 from AxmeAI/docs/readme-add-sections-20260325
docs: add comparison, use cases, pain points, execution model, reliability
2 parents 754ccb5 + 0f907fb commit b7a3ca2

1 file changed

Lines changed: 66 additions & 0 deletions

File tree

README.md

Lines changed: 66 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -19,6 +19,20 @@ AXME is not async RPC. Not a simplified Temporal. Not an agent framework. Not an
1919

2020
---
2121

22+
## Why Ordinary Async Breaks Down
23+
24+
Every team hits the same walls when coordinating work across services, agents, and people:
25+
26+
- Webhooks fail silently or arrive twice
27+
- Retry logic gets scattered across services
28+
- Polling loops add load and hide failures
29+
- Human approvals break automation chains
30+
- No shared execution state exists across time
31+
- Temporal requires determinism constraints and a dedicated platform team
32+
- Workflow engines add complexity when you need a simple approval gate
33+
34+
---
35+
2236
## Before and After
2337

2438
**Without AXME** - polling, webhooks, Redis, and glue code:
@@ -62,6 +76,26 @@ No polling. No webhooks. No Redis. No glue code. The platform handles retries, t
6276

6377
---
6478

79+
## How AXME Compares
80+
81+
| Approach | Lines of code | What you operate |
82+
|---|---|---|
83+
| DIY (webhooks + polling + Redis) | ~200 | Everything: retry logic, state machine, timeout jobs, notification service |
84+
| Temporal | ~80 | Temporal server cluster, workers, determinism constraints |
85+
| AXME | ~15 | Nothing - managed service with protocol-level guarantees |
86+
87+
---
88+
89+
## Who Uses AXME
90+
91+
**Backend teams:** Approval flows, long-running API orchestration, cross-service coordination, replace polling and webhooks.
92+
93+
**AI agent builders:** Human-in-the-loop that waits hours not minutes, multi-agent workflows with checkpoints, production durability across restarts, framework-agnostic (LangGraph, CrewAI, AutoGen, raw Python).
94+
95+
**Platform teams:** One coordination protocol instead of webhook-polling-queue stack, no runtime lock-in, identity and routing built in, simpler than Temporal.
96+
97+
---
98+
6599
## Quick Start
66100

67101
```bash
@@ -170,6 +204,38 @@ How intents reach agents and services:
170204

171205
---
172206

207+
## Execution Model
208+
209+
```
210+
Initiator (SDK/CLI) Handler (your agent)
211+
| |
212+
| POST /v1/intents |
213+
| -----------------------> AXME Cloud |
214+
| (Gateway + Registry) |
215+
| | |
216+
| |--- stream (SSE push)
217+
| |--- poll (agent pulls)
218+
| |--- http (webhook + HMAC)
219+
| |--- inbox (human task queue)
220+
| |--- internal (built-in runtime)
221+
| | |
222+
| GET /v1/intents/{id} | |
223+
| SSE /v1/intents/{id}/events | |
224+
| <----------------------- | <-------------|
225+
| lifecycle events | resume/result
226+
```
227+
228+
---
229+
230+
## Reliability Guarantees
231+
232+
- **At-least-once delivery** with ordered lifecycle events and replay support
233+
- **Idempotency keys** on every write operation - retry-safe by design
234+
- **Five delivery bindings** - choose the right trade-off per agent
235+
- **Disconnect and resume** - initiators can go offline, reconnect later, and pick up where they left off
236+
237+
---
238+
173239
## Human-in-the-Loop
174240

175241
Three paths for human participation:

0 commit comments

Comments
 (0)